Attention-based encoder-decoder networks for state of charge estimation of lithium-ion battery

The state of charge is a significant indicator of the lithium-ion batteries. Most state of charge estimation methods focus on making estimates at the condition of a fixed ambient temperature, However, the ambient temperature in the real-world changes continuously, which poses a significant challenge to accurate estimation. To address this problem, this paper proposes a new attention-based encoder-decoder networks for the state of charge estimation for lithium-ion batteries under complex ambient temperature conditions. First, a bidirectional long short-term memory-based(LSTM) encoder is constructed to obtain the hidden state vector from an input sequence. Second, the hidden state vector from the encoder is input to the sequence pattern attention layer for further processing, after which a new hidden state vector that integrates context and sequence pattern information is obtained. Finally, the new hidden state vector is fed to the decoder to obtain the final result. One public dataset is used to evaluate the performance of the proposed method, and the results of experiments demonstrate that the proposed method outperformed the baseline methods and achieved the best results with MAE within 0.77%.